UPOS (Universal Part-of-Speech) tags are a core component of the Universal Dependencies (UD) project, designed to provide a standardized, fixed set of 17 categories that remain consistent across all human languages. Unlike language-specific systems (XPOS), which reflect the unique morphological intricacies of a single tongue, UPOS focuses on the functional role of a word. By stripping away language-specific "noise," UPOS allows researchers and developers to compare syntactic structures cross-linguistically and facilitates Cross-Lingual Transfer Learning—where an AI model trained on one language (like English) can apply its structural knowledge to another (like Romanian or Korean). It essentially serves as a "Lingua Franca" for computational linguistics, ensuring that a NOUN remains a NOUN whether the underlying grammar is agglutinative, fusional, or analytic.
Try our Chinese UPOS tagging now.
| Group | Tag | Meaning | Example |
|---|---|---|---|
| Open Class | ADJ | Adjective | 大，老，绿，难以理解，第一 |
| ADV | Adverb | 非常，明天，向下，哪里，那里 | |
| INTJ | Interjection | 嘘，哎呀，棒极了，你好 | |
| NOUN | Noun (common) | 女孩，猫，树，空气，美丽 | |
| PROPN | Proper Noun | 玛丽，约翰，伦敦，北约，HBO | |
| VERB | Verb | 跑，跑步，跑着，吃，吃了，吃过 | |
| Closed Class | ADP | Adposition | 在，到，在……期间 |
| AUX | Auxiliary | 是，已经（做），将（做），应该（做） | |
| CONJ | Conjunction | 和，或，但是（遗留标签） | |
| CCONJ | Coordinating Conjunction | 和，或，但是 | |
| SCONJ | Subordinating Conjunction | 如果，当……时，那 | |
| DET | Determiner | 一个，那个（冠词/限定词） | |
| NUM | Numeral | 1, 2017, 一, 七十七, MMXIV | |
| PART | Particle | 的, 不 | |
| PRON | Pronoun | 我, 你, 他, 她, 我自己, 他们自己, 某人 | |
| Other | PUNCT | Punctuation | ., (, ), ?, ] |
| SYM | Symbol | $, %, +, −, :), 🐻 | |
| X | Other / Foreign | sfpksdpsxmsa, ..., foreign words | |
| SPACE | Space | newlines, tabs, extra spaces |
XPOS (Language-Specific Part-of-Speech) tagging offers a much higher level of granularity than the broader UPOS (Universal Part-of-Speech) system. While UPOS provides a standardized set of labels designed to work consistently across every language—ensuring that a NOUN in English is treated similarly to a NOUN in XPOS preserves the unique "linguistic DNA" of a specific language. It is the engine behind complex morphological analysis, allowing a system to distinguish not just that a word is a "Verb," but specifically that it is a "Third-Person, Singular, Past Tense, Passive Voice" verb. By capturing the deep grammatical details that UPOS omits for the sake of universality, XPOS enables the creation of translation tools and parsers that understand the precise inflectional logic of a specific culture and tongue.
In Chinese models, the xpos attribute follows the Penn Chinese Treebank (CTB) tagset. These are specifically designed to capture the unique grammatical structures of the Chinese language, such as aspect markers, localizers, and the distinction between various structural particles. Unlike the English Penn Treebank, the Chinese version treats predicative adjectives (VA) as a sub-type of verb because they function as predicates without a copula.
Try our Chinese XPOS tagging now.
| Category | Tag | Meaning | Example |
|---|---|---|---|
| Verbs | VV | Other verb (main verbs) | 跑 (run), 学习 (study), 喜欢 (like) |
| VA | Predicative adjective | 红 (red), 快 (fast), 高兴 (happy) | |
| VC | Copula (The word "是") | 是 (is/am/are) | |
| VE | Existential "have" (The word "有") | 有 (have/there is), 无 (not have) | |
| Nouns & Pronouns | NN | Common noun | 书 (book), 猫 (cat), 老师 (teacher) |
| NR | Proper noun | 北京 (Beijing), 张三 (Zhang San) | |
| NT | Temporal noun (Time) | 今天 (today), 明年 (next year), 现在 (now) | |
| PN | Pronoun | 我 (I), 你 (you), 他 (he), 这 (this) | |
| Particles & Aspect | DEC | Complementizer/Relative clause "的" | 吃的 (thing to eat), 写的 (written) |
| DEG | Genitive/Associative "的" | 我的 (my), 老师的 (teacher's) | |
| DER | Resultative "得" | 跑得 (run [fast]), 做得 (do [well]) | |
| DEV | Adverbial "地" | 慢慢地 (slowly), 认真地 (earnestly) | |
| AS | Aspect marker | 了 (completed), 着 (ongoing), 过 (past) | |
| SP | Sentence-final particle | 吗 (question), 呢, 吧, 啊 | |
| MSP | Other particle | 所, 以 | |
| Adpositions & Passive | P | Preposition (excluding "把" and "被") | 在 (at), 从 (from), 对 (to/towards) |
| LC | Localizer (Position) | 上 (top), 里 (inside), 中 (middle) | |
| BA | "把" in ba-construction | 把 (object marker) | |
| SB | "被" in short passive (no agent) | 被 (as in "被打了") | |
| LB | "被" in long passive (with agent) | 被 (as in "被他打了") | |
| Modifiers & Conjunctions | AD | Adverb | 也 (also), 很 (very), 不 (not), 已经 (already) |
| JJ | Other noun modifier (non-predicative) | 男 (male), 女 (female), 初等 (elementary) | |
| DT | Determiner | 这 (this), 那 (that), 各 (each), 每 (every) | |
| CC | Coordinating conjunction | 和 (and), 而且 (but also), 或 (or) | |
| CS | Subordinating conjunction | 如果 (if), 虽然 (although) | |
| Numbers & Classifiers | CD | Cardinal number | 一 (one), 100 |
| OD | Ordinal number | 第一 (first), 第二 (second) | |
| M | Measure word / Classifier | 个, 本, 只, 条 | |
| Special & Others | IJ | Interjection | 喂 (hey), 嗯 (um) |
| ON | Onomatopoeia | 哗啦 (splash), 叮当 (jingle) | |
| PU | Punctuation | 。 , ！ , ？ , “ | |
| FW | Foreign word | iPhone, Google | |
| ETC | Et cetera | 等, 等等 |
The DEP (Syntactic Dependency) refers to the specific grammatical relationship between a "child" token and its "head" (parent) token. While primary labels (like nsubj or obj) describe the basic structure, attachments starting with a colon (:) provide fine-grained sub-type information. For instance, while nsubj identifies a subject, :pass refines this to show the subject is being acted upon (Passive Voice). Similarly, :nn (Noun Compound) or :assmod (Associative Modifier) help the parser distinguish between simple modifiers and complex ownership or compound relationships, allowing for a much deeper "logical" understanding of the sentence.
| Category | Label | Meaning | Example (Token in bold) |
|---|---|---|---|
| Core Arguments | nsubj | Nominal subject | 埃隆吃东西。 |
| csubj | Clausal subject | 他所做的事是错的。 | |
| obj | Direct object | 我看着月亮。 | |
| iobj | Indirect object | 她给了我一件礼物。 | |
| ccomp | Clausal complement (finite) | 他说他累了。 | |
| xcomp | Open clausal complement | 我想去。 | |
| Non-Core Dependents | obl | Oblique nominal | 他坐在椅子上。 |
| vocative | Vocative | 约翰，过来！ | |
| expl | Expletive | 这儿有一只猫。 | |
| dislocated | Dislocated element | 那个男人，我认识他。 | |
| advcl | Adverbial clause modifier | 在他到达之后我离开了。 | |
| advmod | Adverbial modifier | 跑得快。 | |
| discourse | Discourse element | 嗯，我不确定。 | |
| aux | Auxiliary | 我能看见。 | |
| cop | Copula | 她很高兴。 | |
| mark | Subordinating marker | 我知道你知道。 | |
| Nominal Dependents | nmod | Nominal modifier | 汽车的门。 |
| appos | Appositional modifier | 山姆，我的朋友。 | |
| nummod | Numeric modifier | 七天。 | |
| acl | Adjectival clause | 获胜的计划。 | |
| amod | Adjectival modifier | 蓝色的天空。 | |
| det | Determiner | 终结。 | |
| case | Case marking | 法国的国王。 | |
| fixed | Fixed multiword expression | 尽管那样。 | |
| flat | Flat multiword name | 纽约市。 | |
| compound | Compound noun | 电话亭。 | |
| list | List element | 电话、钥匙、钱包。 | |
| Coordination | conj | Conjunct | 面包和黄油。 |
| cc | Coordinating conjunction | 面包和黄油。 | |
| Special Labels | aux:pass | Passive auxiliary | 它被偷了。 |
| punct | Punctuation | 你好！ | |
| dep | Unspecified dependency | （用于未知连接） | |
| ROOT | Root of the sentence | 我吃过午饭。 |
| Attachment | Full Name | Explanation | Example |
|---|---|---|---|
| :pass | Passive | Indicates a relationship in a passive voice construction. | nsubj:pass (窗户被打破了) |
| :nn | Noun Compound | Indicates that a noun is modifying another noun in a compound structure. | compound:nn (电话充电器) |
| :prep | Prepositional | Refines a modifier governed specifically by a preposition. | nmod:prep (垫子上的猫) |
| :assmod | Associative Modifier | Common in Romanian/Baltic languages; shows nouns modifying other nouns. | nmod:assmod (我父亲的车) |
| :poss | Possessive | Indicates ownership or a possessive relationship. | nmod:poss (我的狗，约翰的帽子) |
| :relcl | Relative Clause | Identifies a clause that modifies a noun phrase. | acl:relcl (我读的书) |
| :tmod | Temporal Modifier | A modifier specifically describing time or duration. | nmod:tmod (我会在周二离开) |
| :prt | Particle | Used for phrasal verb particles. | compound:prt (放弃，关闭) |
| :rcomp | Relative Complement | Used for complements of relative clauses (common in Dutch). | advcl:rcomp (离开的男人) |
| :flat | Flat Modifier | Used for multi-word expressions that don't have a clear internal head. | flat:name (奥巴马总统) |
NER (Named Entity Recognition) is a Natural Language Processing (NLP) task that automatically identifies and categorizes key information (entities) in a text into predefined classes. In spaCy, the statistical model "looks" at the context of a word to determine if it refers to a person, an organization, a monetary value, or a specific date. This is crucial for extracting structured data from unstructured text, such as finding all the company names mentioned in a news article or identifying the dates of events in a history book.
Comparison Note: GPE vs. LOC
Determining whether a place is a GPE or a LOC depends on its political nature:
GPE (Geopolitical Entity): If the location has a government, specific laws, or human-defined administrative borders, it is labeled as a GPE. Examples include Seoul, Germany, the United Kingdom, and California.
LOC (Location): If the place is a natural physical feature or a broad geographic region without a singular governing body, it is labeled as a LOC. Examples include the Alps, the Pacific Ocean, the Middle East, and Mount Everest.
| Label | Meaning | Example |
|---|---|---|
| 🌍 GPE | Geopolitical entity (countries, cities, states) | 中国，北京，法国，加利福尼亚 |
| 🏔️ LOC | Non-political location (mountains, rivers) | 太平洋，珠穆朗玛峰，阿尔卑斯山脉 |
| 🏢 FAC | Facility (buildings, airports, highways) | 金门大桥，肯尼迪机场，哈利法塔 |
| 👤 PERSON | People (real or fictional) | 埃隆·马斯克，哈利·波特，艾伦·图灵 |
| 🚩 NORP | Nationalities, religious or political groups | 中国人，佛教徒，民主党人，日本人 |
| 🏢 ORG | Organizations (companies, institutions) | 谷歌，联合国，苹果公司，国际足联 |
| 📅 DATE | Absolute or relative dates | 2026年7月4日，昨天，下周 |
| ⌚ TIME | Times smaller than a day | 上午9:30，日落，十分钟 |
| 🎊 EVENT | Named events (wars, festivals) | 第二次世界大战，科切拉音乐节，奥林匹克运动会 |
| 💰 MONEY | Monetary values, including unit | 100美元，500万欧元，50英镑 |
| ‱ PERCENT | Percentage, including "%" | 20%，百分之八十，0.5% |
| ⚖️ QUANTITY | Measurements (weight, distance) | 5公里，100磅，30平方米 |
| 🔢 ORDINAL | "First", "second", etc. | 第一，第二，第九 |
| 🔢 CARDINAL | Numbers not classified elsewhere | 10，一千，三 |
| 📦 PRODUCT | Objects, vehicles, foods, etc. (not services) | iPhone，特斯拉Model S，可口可乐 |
| 🎨 WORK_OF_ART | Titles of books, songs, etc. | 蒙娜丽莎，波西米亚狂想曲，哈姆雷特 |
| 📜 LAW | Named legal documents | 宪法，凡尔赛条约 |
| 🗣️ LANGUAGE | Named languages | 中文，Python，英语 |
如果我们处理短语“谷歌总部位于加利福尼亚” (Google is based in California)，各层分析如下：
Lemma (词干/原形): “谷歌”，“总部”，“位于”，“加利福尼亚”
UPOS (通用词性): “PROPN(专有名词)”，“NOUN(名词)”，“VERB(动词)”，“PROPN(专有名词)”
XPOS (具体词性/宾州树库): “NR(专有名词)”，“NN(常见名词)”，“VV(其他动词)”，“NR(专有名词)”
DEP (依存关系): “谷歌”是“总部”的 nmod:assmod (关联修饰语)；“总部”是“位于”的 nsubj (名词主语)；“位于”是整个句子的核心动词，即 Root (核心节点)；“加利福尼亚”是“位于”的 obj (宾语)。
NER (命名实体识别): “谷歌”是 🏢 ORG (组织机构)，“加利福尼亚”是 🌍 GPE (地缘政治实体)。
Arabic -
Catalan -
Chinese -
Classical Chinese -
Croatian -
Danish -
Dutch -
English -
Filipino -
Finnish -
French -
German -
Greek -
Hebrew -
Hindi -
Italian -
Indonesian -
Japanese -
Korean -
Latin -
Lithuanian -
Macedonian -
Norwegian -
Polish -
Portuguese -
Romanian -
Russian -
Slovenian -
Sanskrit -
Spanish -
Swedish -
Tamil -
Thai -
Ukrainian -
Vietnamese
|
|
|
© Stars21 - All Rights Reserved
|
|||||